Chapter 2: Control Data Collection › Explain the Primary Performance Agent › How Primary Data Collection Is Synchronized in a Cluster

How Primary Data Collection Is Synchronized in a Cluster

The primary data collection process, controlled by the CPD collection definition, gathers information from the system's internal data structures. Some data collection is event driven, some is sampled periodically, and some is sampled at the end of each interval. At the end of each interval, all the information is time-stamped and written to the daily data file with a file type of CPD. All intervals have a fixed duration of two minutes. The time-stamp identifies each interval's data records.

Because each node in a cluster has its own clock and is set independently, there tends to be some variance, great or small, between the clocks. To make an analysis of the data meaningful and effective, the intervals recorded on each node must represent the identical period of real time.

To accomplish this synchronization, the first primary data collection process to start in a cluster initializes a logical clock with the node's system time. Using the Lock Manager, that clock is distributed to all other data collection processes as they start in a cluster. This synchronization strategy guarantees that data sampled independently on separate nodes in a cluster correlate accurately.